Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Auction-based Marketplace for Model Trading in Federated Learning

Yue Cui, Liuyi Yao, Yaliang Li, Ziqian Chen, Bolin Ding, Xiaofang Zhou

TL;DR

The paper reframes federated learning as a model-trading marketplace in cross-silo FL, introducing performance-gain based pricing and an RL-driven allocation framework to maximize trading volume while preserving fairness and accuracy. It establishes theoretical mechanisms to encourage truthful bidding, enforces sale-copy limits and selective seller authorization to sustain market dynamics, and uses an A2C-based solver to optimize allocations across evolving market states. Empirical results on MNIST, FMNIST, FEMNIST, and CIFAR-10 show increased trading volumes and competitive downstream accuracy against a GSP baseline, with reward shaping improving fairness for bottom-tier participants. Overall, the work advances liquidity and incentive compatibility in FL by integrating auction theory with reinforcement learning for market operations.

Abstract

Federated learning (FL) is increasingly recognized for its efficacy in training models using locally distributed data. However, the proper valuation of shared data in this collaborative process remains insufficiently addressed. In this work, we frame FL as a marketplace of models, where clients act as both buyers and sellers, engaging in model trading. This FL market allows clients to gain monetary reward by selling their own models and improve local model performance through the purchase of others' models. We propose an auction-based solution to ensure proper pricing based on performance gain. Incentive mechanisms are designed to encourage clients to truthfully reveal their model valuations. Furthermore, we introduce a reinforcement learning (RL) framework for marketing operations, aiming to achieve maximum trading volumes under the dynamic and evolving market status. Experimental results on four datasets demonstrate that the proposed FL market can achieve high trading revenue and fair downstream task accuracy.

An Auction-based Marketplace for Model Trading in Federated Learning

TL;DR

The paper reframes federated learning as a model-trading marketplace in cross-silo FL, introducing performance-gain based pricing and an RL-driven allocation framework to maximize trading volume while preserving fairness and accuracy. It establishes theoretical mechanisms to encourage truthful bidding, enforces sale-copy limits and selective seller authorization to sustain market dynamics, and uses an A2C-based solver to optimize allocations across evolving market states. Empirical results on MNIST, FMNIST, FEMNIST, and CIFAR-10 show increased trading volumes and competitive downstream accuracy against a GSP baseline, with reward shaping improving fairness for bottom-tier participants. Overall, the work advances liquidity and incentive compatibility in FL by integrating auction theory with reinforcement learning for market operations.

Abstract

Federated learning (FL) is increasingly recognized for its efficacy in training models using locally distributed data. However, the proper valuation of shared data in this collaborative process remains insufficiently addressed. In this work, we frame FL as a marketplace of models, where clients act as both buyers and sellers, engaging in model trading. This FL market allows clients to gain monetary reward by selling their own models and improve local model performance through the purchase of others' models. We propose an auction-based solution to ensure proper pricing based on performance gain. Incentive mechanisms are designed to encourage clients to truthfully reveal their model valuations. Furthermore, we introduce a reinforcement learning (RL) framework for marketing operations, aiming to achieve maximum trading volumes under the dynamic and evolving market status. Experimental results on four datasets demonstrate that the proposed FL market can achieve high trading revenue and fair downstream task accuracy.
Paper Structure (30 sections, 4 theorems, 20 equations, 6 figures, 5 tables, 1 algorithm)

This paper contains 30 sections, 4 theorems, 20 equations, 6 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Preliminary
  3. Methodology
  4. Market Design
  5. Principles to Incentives True Valuation Revealing
  6. Reinforcement Learning Based Auction Allocation
  7. Experiment
  8. Experiment Setup
  9. Main Results
  10. The FL Market's Robustness to Bidding Strategies
  11. Social Fairness Analysis
  12. Related Work
  13. Federated Learning
  14. Incentive Mechanism Design in FL
  15. Conclusion and Future Work
  16. ...and 15 more sections

Key Result

Theorem 3.1

When each model seller has no limits on the number of the sold model copies, no model buyer will have the incentive to submit a bid equaling its predicted monetary utility unless there exists a constraint on the number of model copies:

Figures (6)

  • Figure 1: Pipeline of the auction-based FL market.
  • Figure 2: Compare with the baseline on Trading Volumes and downstream task Accuracy on MNIST.
  • Figure 3: Effect of the reward shaping mechanism.
  • Figure 4: The reinforcement learning based solution for auction-based FL market.
  • Figure 5: Compare with the baseline on Trading Volumes and downstream task Accuracy.
  • ...and 1 more figures

Theorems & Definitions (11)

  • Definition 2.1
  • Definition 2.2
  • Definition 3.1
  • Definition 3.2
  • Definition 3.3
  • Theorem 3.1
  • Theorem 3.2
  • Theorem 2.1: Theorem \ref{['th:copy']} Restate
  • proof
  • Theorem 2.2: Theorem \ref{['th:curse']} Restate
  • ...and 1 more